jastrachan | 26 Nov 10:03 2004
Picon

variable declaration syntax: use cases and thoughts

We discussed this at the GC1 (Groovy Conference#1) and we used the 
token 'def' to represent variable/field declarations.

Firstly in Java variable declarations are always different to 
assignments as a type is mandatory. So we're already used in the Java 
world to explicit variable declarations. The issue for groovy is, if 
the type is optional, there's no difference between them from a syntax 
perspective, which causes some issues...

1. Catching common typos

foot = 1
if (bar) foo = 2  // the user mistyped the variable name
print foot

2. Declarations without value assignment

Since we're hoping to use the no-dumb-expression handling to catch 
types and bad newlines, statements such as the following are bad

     x
     +x

So we can't just define variables with 1 token. We need a declaration 
token (or to disallow variable declarations without a value - i.e. use 
regular assigment x = 1).

e.g.

class Foo {
     x
}

is a bit wierd; our mental grammar models makes this feel wrong. So 
with a variable/field declaration token this feels better

class Foo {
    x = 1
}

or

class Foo {
    var x
}

It must be said that in interactive shells, folks are used to typing 
things like

     x = 123

and not having to make explicit variable declarations; so I think the 
requirement for variable declaration syntax is mostly for defining 
fields and variables inside method bodies.

There's been much debate over def / any / let / var etc. Many people 
have been plain confused by 'any' which despite its simplicity, I'm now 
thinking we should avoid - we probably just need a token associated 
with declaring a field/variable. I'm currently liking something from 
JavaScript - using 'var' is my current favourite.

someMethod() {
     var foot = 123
     if (bar) {
	foo = 4  // BAD compile error unknown variable
     }
     return foot
}

Then using 'def' for defining methods/functions.

Coming from Java, the use of 'var' (or any new token) is a little 
strange at first; but then we're used to typing some token anyways in 
Java for declarations.

I've grappled with this issue for some time and I'm now pretty 
convinced we need some simple token for those wishing to define a 
variable with no static type; so far I'm now leaning towards 'var'. 
Thoughts?

James
-------
http://radio.weblogs.com/0112098/


Gmane